The question comes up in almost every psychology, nursing and health sciences thesis: "Can I treat my Likert scale data as if they were continuous?" The answer is not unique and depends on what you are going to do with them. This guide covers the options, the criteria for choosing, and the concrete steps in SPSS.
If what you need is the item-by-item analysis (discrimination, item-total correlation, alpha-if-item-deleted), the item analysis calculator computes it directly without building it by hand in SPSS.
The ordinal vs interval debate: what it is really about
A 5- or 7-point Likert scale produces ordinal data: we know that "strongly agree" is more than "agree", but we do not know whether the distance between each category is equal. Technically, that means the mean is not meaningful because it assumes the intervals are equal.
However, decades of psychometric research have shown that, in practice, treating scores from scales with 5 or more categories as continuous produces robust results when the distribution is not highly skewed. This is what most studies published in Q1 psychology journals do and what reviewers expect to see when you have a validated instrument with a known factor structure.
The current consensus (Rhemtulla, Brosseau-Liard and Savalei, 2012; Norman, 2010) can be summarised as follows:
- With 5 or more categories and approximately symmetric distribution: you can use parametric analyses (t-test, ANOVA, regression, Pearson correlation, CFA with ML or MLR estimator).
- With 4 or fewer categories, highly skewed distribution, or confirmatory factor analysis: use ordinal-specific methods (polychoric correlations, WLSMV estimator in CFA).
- For individual items (not summed) with few participants: consider non-parametric tests.
First: define clearly what you are analysing
There are two distinct objects of analysis and it helps to keep them separate:
- Individual items: each 1-to-5 response is ordinal. If you want to describe how the sample responds to a specific item, the median and interquartile range are more appropriate than the mean. For group comparisons on a single item, use Mann-Whitney or Kruskal-Wallis.
- Total or subscale score (sum or mean of several items): this composite score behaves more like a continuous variable, especially with 6 or more items. The mean of 10 five-point items has a distribution that approximates normality by the central limit theorem. Parametric analyses are defensible here.
Steps in SPSS for descriptive analysis
- Open the data file and identify which variables are questionnaire items.
- Calculate the total score: Transform > Compute Variable. Sum the items (remember to reverse-score inverted items first using Transform > Recode into Different Variables).
- For the total score: Analyze > Descriptive Statistics > Descriptives (mean, SD, skewness, kurtosis).
- For individual items: Analyze > Descriptive Statistics > Frequencies (frequencies, percentages, median).
- Check normality of the total score if you plan to use ANOVA or regression: Analyze > Descriptive Statistics > Explore > Plots > Normality plots with tests.
Group comparisons in SPSS
If you want to compare means of the total score between two groups (for example, men vs women on a burnout scale):
- Parametric option (recommended if N is 30 or more per group and the distribution is acceptable): Analyze > Compare Means > Independent-Samples T Test. Report t, degrees of freedom, p and Cohen's d (which SPSS does not calculate directly; use the effect size calculator).
- Non-parametric option (if the distribution is highly skewed or N is small): Analyze > Nonparametric Tests > Legacy Dialogs > 2 Independent Samples > Mann-Whitney U. Report U, Z and p; the effect size is r = Z / sqrt(N).
For more than two groups, the parametric equivalent is ANOVA (Analyze > Compare Means > One-Way ANOVA) and the non-parametric equivalent is Kruskal-Wallis (Analyze > Nonparametric Tests > Legacy Dialogs > K Independent Samples). The complete ANOVA tutorial is in the guide to SPSS for your thesis.
Correlation between Likert scales
For total scores from two scales (for example, anxiety and depression): Pearson correlation is appropriate with composite scores. If the variables are very non-normal or you have few data points, use Spearman. In SPSS: Analyze > Correlate > Bivariate (tick Pearson or Spearman as appropriate).
Do not use Pearson between two individual Likert items without further consideration: that correlation underestimates the true relationship because categorisation truncates the variance. For matrices of individual items, polychoric correlations (which SPSS does not compute natively; use R or jamovi with the appropriate module) are more appropriate as a basis for CFA.
Validating the structure with factor analysis
If the questionnaire has not been validated before in your population, you need to verify that the items form the dimensions you expect. Confirmatory factor analysis (CFA) is the standard for instruments with a prior theoretical structure; exploratory factor analysis (EFA) is for when you have no clear hypotheses about the structure. The EFA guide and the CFA in APA 7 guide cover both in detail; for the write-up, see how to report an exploratory factor analysis in APA 7.
An important technical note: if you run CFA with Likert items that have few categories (4 or fewer) or a skewed distribution, the classic ML estimator is not ideal. Use polychoric correlations with the WLSMV estimator (available in R with lavaan or in Mplus). With 5 or more categories and ML, add the argument estimator = "MLR" in lavaan for robust standard error correction.
Reliability: Cronbach's alpha and McDonald's omega
Once the structure is defined, check that the items within each dimension are internally consistent. In SPSS: Analyze > Scale > Reliability Analysis, choose Cronbach's alpha. The commonly used reference values are ≥ .70 for group-level use and ≥ .90 for individual diagnostic use. McDonald's omega is a preferable alternative to alpha when factor loadings are not homogeneous; the omega calculator computes it directly.
How to report in APA 7
For group comparison: t(58) = 3.14, p = .003, d = 0.81. For correlation: r(88) = .43, p < .001. For reliability: Cronbach's alpha = .87. Always specify whether you used the total score or individual items, the analysis method (parametric or non-parametric) and the justification. More formats in the APA 7 reporting guide.
Do you have a Likert scale and are not sure which analysis to run?
I am a PhD in Psychology and I analyse your scale from start to finish: factor structure, reliability, group comparisons and APA reporting. Psychology, nursing, physiotherapy or medicine. Free initial diagnosis.
Book free diagnosis →Frequently asked questions
Can you calculate the mean with Likert scale data?
With 5 or more categories and approximately symmetric distribution, yes. Psychometric research shows that treating scores from scales with 5 or more categories as continuous produces robust results. For individual items with few categories or a highly skewed distribution, the median and non-parametric tests are more appropriate.
When should I use a t-test instead of Mann-Whitney with Likert data?
Use the t-test when analysing the total scale score (sum of several items), you have 30 or more participants per group, and the distribution is not highly skewed. Use Mann-Whitney when analysing individual items, the sample is small, or the distribution is clearly asymmetric.
Do I need to run factor analysis with my Likert scale?
It depends on whether the questionnaire has been validated previously in your population. If you are using an instrument with a known and published factor structure, you are not obliged to redo it. If your sample is different (clinical vs general population, different language), a confirmatory analysis is recommended to verify that the structure holds.
How do I calculate the total score of a Likert scale in SPSS?
Go to Transform > Compute Variable. Enter the name for the new variable and in the expression box sum the items: item1 + item2 + item3. If there are reversed items, first recode them using Transform > Recode into Different Variables before summing.